Detección de la señalización de tránsito vertical con redes neuronales convolucionales basadas en bloques residuales

Abstract

The objective of the present work is to train a neural network capable of detecting vertical traffic signaling and classify it using residual blocks, as they allow deeper neural networks. The methodology used for the development of the neural network comprises four phases: neural network definition, training, utilization, and maintenance of the neural network. For the development of theneural network there are two datasets, the first is of German origin, consists of 50,000 images and is widely used for the classification of traffic signs; and the second of Bolivian origin, which has 9,548 road images. The percentage of efficiency of theneural network no. 1 with the GTSRB dataset is high, obtaining a value of 94.36%, it also includes high values in the classification report, otherwise, it happens with the Bolivia dataset because the dataset is unbalanced.